You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
yt-dlp/yt_dlp/extractor/parler.py

92 lines
3.7 KiB
Python

import functools
from .common import InfoExtractor
from .youtube import YoutubeIE
from ..utils import (
clean_html,
int_or_none,
strip_or_none,
traverse_obj,
unified_timestamp,
urljoin,
)
class ParlerIE(InfoExtractor):
IE_DESC = 'Posts on parler.com'
_VALID_URL = r'https?://parler\.com/feed/(?P<id>[0-9a-f]{8}-(?:[0-9a-f]{4}-){3}[0-9a-f]{12})'
_TESTS = [
{
'url': 'https://parler.com/feed/df79fdba-07cc-48fe-b085-3293897520d7',
'md5': '16e0f447bf186bb3cf64de5bbbf4d22d',
'info_dict': {
'id': 'df79fdba-07cc-48fe-b085-3293897520d7',
'ext': 'mp4',
'thumbnail': 'https://bl-images.parler.com/videos/6ce7cdf3-a27a-4d72-bf9c-d3e17ce39a66/thumbnail.jpeg',
'title': 'Parler video #df79fdba-07cc-48fe-b085-3293897520d7',
'description': 'md5:6f220bde2df4a97cbb89ac11f1fd8197',
'timestamp': 1659785481,
'upload_date': '20220806',
'uploader': 'Tulsi Gabbard',
'uploader_id': 'TulsiGabbard',
'uploader_url': 'https://parler.com/TulsiGabbard',
'view_count': int,
'comment_count': int,
'repost_count': int,
},
},
{
'url': 'https://parler.com/feed/f23b85c1-6558-470f-b9ff-02c145f28da5',
'md5': 'eaba1ff4a10fe281f5ce74e930ab2cb4',
'info_dict': {
'id': 'r5vkSaz8PxQ',
'ext': 'mp4',
'live_status': 'not_live',
'comment_count': int,
'duration': 1267,
'like_count': int,
'channel_follower_count': int,
'channel_id': 'UCox6YeMSY1PQInbCtTaZj_w',
'upload_date': '20220716',
'thumbnail': 'https://i.ytimg.com/vi/r5vkSaz8PxQ/maxresdefault.jpg',
'tags': 'count:17',
'availability': 'public',
'categories': ['Entertainment'],
'playable_in_embed': True,
'channel': 'Who Knows What! With Mahesh & Friends',
'title': 'Tom MacDonald Names Reaction',
'uploader': 'Who Knows What! With Mahesh & Friends',
'uploader_id': '@maheshchookolingo',
'age_limit': 0,
'description': 'md5:33c21f0d35ae6dc2edf3007d6696baea',
'channel_url': 'https://www.youtube.com/channel/UCox6YeMSY1PQInbCtTaZj_w',
'view_count': int,
'uploader_url': 'http://www.youtube.com/@maheshchookolingo',
},
},
]
def _real_extract(self, url):
video_id = self._match_id(url)
data = self._download_json(f'https://api.parler.com/v0/public/parleys/{video_id}',
video_id)['data']
if data.get('link'):
return self.url_result(data['link'], YoutubeIE)
return {
'id': video_id,
'title': strip_or_none(data.get('title')) or '',
**traverse_obj(data, {
'url': ('video', 'videoSrc'),
'thumbnail': ('video', 'thumbnailUrl'),
'description': ('body', {clean_html}),
'timestamp': ('date_created', {unified_timestamp}),
'uploader': ('user', 'name', {strip_or_none}),
'uploader_id': ('user', 'username', {str}),
'uploader_url': ('user', 'username', {functools.partial(urljoin, 'https://parler.com/')}),
'view_count': ('views', {int_or_none}),
'comment_count': ('total_comments', {int_or_none}),
'repost_count': ('echos', {int_or_none}),
})
}